In October 2026, the Collège des médecins du Québec (CMQ) published two documents on choosing AI tools and maintaining medical records. For a clinic, they raise a practical question: when should the use of an AI tool be documented in the patient’s record? CMQ publications (French)
The answer starts with what the tool actually does. Drafting a note, receiving information about a person and suggesting an investigation are different uses, even when they sit inside the same software.
Identify what needs to be documented
In its guidance on medical records, the CMQ says it is essential to document the use of an AI system when it has access to information that can identify a patient directly or indirectly, or contributes to clinical reasoning for that patient. Consulting a system for general information, without providing information about a particular patient, does not need to be documented on that basis.
The record should include the system’s name, the date, the main information it provides and, where required, consent. If a physician rejects a plausible recommendation that could have a significant impact, the CMQ recommends recording that decision. Depending on the potential impact, the reasons for disagreeing may also be relevant. AI and medical records (French)
For the team, I would start by listing the functions actually in use. One line per function is enough to open the discussion: who uses it, what data goes in, what comes out and where that output enters the workflow. The software’s brand name does not answer those questions.
Map the AI scribe’s workflow
The CMQ points to separate guidance for notes produced by an AI scribe. It requires explicit consent before use and a check of the note’s accuracy before signing, and recommends identifying the software in the note. The CMQ also requires physicians to choose a scribe certified by Santé Québec’s certification bureau. Physicians using a scribe outside an institution must ensure that a privacy impact assessment (PIA, or ÉFVP in French) has been completed for their workplace. AI scribes (French)
A clinic can turn these points into a simple workflow: the discussion with the patient, the draft, the physician’s review, signature and access to the final note. At each handoff, the team needs to know who takes over and how an error gets corrected.
I would test that workflow with fictional scenarios before changing its settings. Where does the tool’s name appear? Who catches a sentence that should not have been added? What does the next professional see while the note is still awaiting review? These questions help test the process without bringing patient data into a training exercise.
Make review part of everyday work
A “reviewed” checkbox is more useful when it comes with clear instructions. The medical team can define what to check, how to correct an omission and where to escalate an output that remains uncertain. The operations lead can then check whether the software, access permissions and time allocated actually support that work.
I would keep two records separate. The medical record holds information relevant to care. A restricted operational log tracks workflow problems, who is responsible for them and the next action, while limiting personal information to what is necessary. It should not become a parallel collection of patient conversations or records.
Review the process when the tool changes
The CMQ’s guidance on choosing AI tools calls for monitoring over time, including when versions or the context of use change. It also addresses identifying problems, taking corrective action, and adjusting or suspending use when the risks warrant it. Choosing and monitoring AI tools (French)
To make that follow-up workable, I would suggest a short review with a named owner: what changed, which steps are affected and what needs checking before continuing? The frequency and criteria should reflect the risk and be defined with the medical team.
The first useful deliverable can fit on one page: the uses, documentation decisions, responsibilities and points to revisit. It gives the clinic a concrete way to discuss its workflow with physicians, privacy leads and software providers.
Teché helps teams assess their workflows and provides practical training. To clarify these steps in your clinic, let’s discuss your training and workflow needs.
Guidance on medical records, tool selection and AI scribes rechecked on October 30, 2026. This article discusses workflow organization based on the CMQ’s professional guidance. It does not replace medical, legal or privacy analysis suited to your setting.
